RadeonSI Got Another Optimization After The 13.0 Branching

[liam]

In my experience, Mesa developers are highly resistant to innovations which would reduce the amount of work needed.

There does not exist a different way except for the way of letting them implement it at their own pace.

It's related to how quickly a person is able to learn new concepts and start using new concepts.

Can you relate your experience and cite cases where they demonstrate that they won't include a feature because they don't understand it?

[pal666]

Can you relate your experience and cite cases where they demonstrate that they won't include a feature because they don't understand it?

well, it is obvious that you can't maintain what you can't understand. so the one willing to bring innovation has to either maintain it forever or educate everyone involved or shut up

[dungeon]

I guess saint row: the third should also benefit from the patch. As it is, the game is so slow that it's unplayable on my machine (HD7970)

That also wanna threaded GL... basically i would dare to say _ANY_ eONed "port" crying for in driver threaded optimization boostero

[dungeon]

In similar way as wine crying for csmt or nine... it is CPU cap everywhere

[marek]

Is there any incentive to get multithreaded GL working ? I assume it's quite a lot of work to get that right ?

It's partially done actually. Intel did it in 2012-2013, but they stopped. There is just some work needed to wire up it with Gallium. I guess it won't be very fast initially, but it can be optimized after that.

[eydee]

Having better multithreaded optimizations in Mesa would only solve part of the problem. It could help move closer to nvidia blob performance, but not move closer to native windows performance, which (in my opinion) should be the goal of anyone working on the game and related software. Another issue is the linux engine itself, the engine written by VP, which has some limitations when it comes to threading. In a github thread, one of the game developers shares some details in a conversation with a nouveau developer: https://github.com/virtual-programmi...ment-210793972

[dungeon]

Having better multithreaded optimizations in Mesa would only solve part of the problem. It could help move closer to nvidia blob performance, but not move closer to native windows performance...

How is that even possibile , if you get better performance isn't it that you are closer to both?

It is mission impossibile to have exact same performance of Direct3D games when ported to OpenGL... esspecially with thoe wrappers, you can always count how much you lose something in translation... I guess the opposite is true too, it is just that probably no one does OpenGL wrappering to DX

[liam]

well, it is obvious that you can't maintain what you can't understand. so the one willing to bring innovation has to either maintain it forever or educate everyone involved or shut up

Um, OK? I mean, yeah, you're right, but that doesn't really address my question

[marek]

The two major things radeonsi needs for better user experience are multithreading and shader cache. And please atomsymbol, don't confuse innovations with changing coding conventions. While Mesa doesn't change its conventions very often, there are innovative things happening all the time.

[mazumoto]

It's partially done actually. Intel did it in 2012-2013, but they stopped. There is just some work needed to wire up it with Gallium. I guess it won't be very fast initially, but it can be optimized after that.

Thanks for the reply, marek :-) I take that as good news ... I thought it'd be quite difficult to introduce multithreading to Mesa, but if some parts are done already, it might not be too bad to get the basics working.